Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Vol. 1(5), 17-20, June (2013) Res. J. Mathematical and Statistical Sci.
Introduction
In time series econometrics, nonstationarity test is essential for analyzing the behavior of time series and for advance research.
Usually this can be tested by different unit root tests such as Dickey-Fuller (DF) test, Augmented Dickey-Fuller (ADF) test,
Phillips-Perron (PP) test and Kwiatkowski- Phillips-Schmidt-Shin (KPSS) test discussed by Dickey, D. A., and W. A. Fuller1, 2, D.
Kwiatkowski, P. C. B. Phillips, P. Schmidt, and Y. Shin3., Dickey, D. A., Hasza, D. P., and Fuller, W. A.4 , Dickey D. A. and
Gonzalez-Farias, G.5 , Majumder, A. K.6, and Enders, W.7. Almost all the unit root tests as well as the estimated model suffer from
sign and boundary problems of the parameters. According to the assumption of the Dickey-Fuller test, < 1 or 2 < < 0
of the time series models, such as y t = y t 1 + u t . Any estimated value of less than 2 or greater than 0 may results in invalid
model. This invalid model cannot be used for making decision regarding nonstationarity. To overcome this situation it is necessary
to make suitable restrictions on the parameters, which is greater than zero. Hence, the usual DF test for testing unit root is not
always suitable and we need to develop an appropriate one-sided hypothesis test of unit root. But, very few literatures are available
on this issue. In this paper, the main attempt to find a suitable estimation and testing procedure for unit root with constraint.
Restricted Estimation and hypothesis testing: In testing restricted alternatives two stage estimation methods are required in
distance based techniques; in the first stage we estimate the parameter(s) by usual method, such as, least square method, maximum
likelihood method, etc. In the second stage we estimate the constraint parameter(s) by optimizing under the restriction.
All of the hypotheses are restricted. But the hypothesis (5) is unrestricted. In case of all the hypotheses except the hypothesis (5)
the usual two-sided test cannot be applied. Hence we have to develop one-sided or partially one-sided testing procedures.
There are different methods to develop such testing procedures. Among them we are mainly concerned about distance-based
approach because our testing approach is developed using this approach whose null hypothesis distribution follows a mixture of
corresponding two-sided distribution. The following paragraph illustrates aforesaid problems. For testing strictly one-sided or
partially one-sided alternative we make decision about accepting or rejecting the null hypothesis depending on whether, in some
sense, the estimated parameter under test is closer to the null value or to an alternative value. We call this approach to hypothesis
test a distance-based approach.
The usual Dickey-Fuller (DF) test: The usual Dickey-Fuller test examines the condition that the model has a unit root and
differencing helps to remove this unit root. Dickey-Fuller suggested that under the null hypothesis the estimated coefficient of
y t 1 in the model y t = yt 1 + t follows the tau ( ) statistic, must be compared to critical values tabulated by Dickey-Fuller.
In the above three models, the null hypothesis is that = 0 implies that = 1 . That is the time series is non-stationary it means
that it has unit root. The alternative hypothesis is that is less than 0 implies that the time series is stationary and no decision is
made on the basis of alternative hypothesis.
Proposed Dickey-Fuller (DF) test: The DF test can be applied to the three models
i. y
t = yt 1 + ut , ii. yt = + yt 1 + ut , iii. yt = + t + yt 1 + ut
All of the models are valid when < 1 i.e., 2 < < 0 . So models are invalid when values are different from values
are specified above. In constructing DF test ignores this restriction. Ignorance of these two restrictions may result in three different
problems: i. Estimated parameters values will be overestimated or under estimated, ii. At the same time the test statistic based on
this estimates loss power due to ignoring of these restrictions. iii. Very often ignorance of this sign problems result in invalid
estimated models.
Decisions based on these models are quite unacceptable. In the following table we report the proportion of time, we may ignorance
of sign problem provide invalid estimated models by Monte Carlo Simulation.
We generated each model under H 0 100 times with sample sizes 20, 50, 100, 200, 500 .We assumes that the true values of
and for the last two models as 1 and 2, respectively. , we observe that estimation are invariant with the choice of and .
The proportion of time usual and our proposed (optimized) estimation procedure provides invalid estimated equations.
Table-1
The proportion of time the estimated values are invalid for the usual and optimized procedure
Proportion of Proportion of Proportion of
Sample True value True value
Model
size n of time value of time value is time optimized
is positive positive value is positive
20 37 35
50 35 37
yt = yt 1 + ut 100 35 28
200 32 30
500 30 22
20 30 33
50 38 33
All of the values
yt = + yt 1 + ut 100 0 42 -2 34
are zero
200 55 31
500 38 40
20 87 43
50 75 41
yt = + t + yt 1 + ut 100 55 41
200 34 32
500 45 32
Since the parameters < 0, < < < , we estimate the model by minimizing the error sum of squares for three models
are:
T 2 T 2 T 2 T 2 T 2 T 2
Sometimes the decision based invalid and valid proposed (optimized) procedure may be similar but the estimated coefficients are
likely to be different. The following simulation results support our argument.
Results
The usual DF test is not totally suitable for testing the hypothesis mentioned above. This is because the hypothesis is strictly one-
sided. For testing this hypothesis, we are interested to develop a testing procedure using distance-based approach called distance-
based DF test. The test statistic of our proposed DF test follows weighted mixture of distribution whereas the test statistic of
usual DF test follows distribution.
Conclusion
Most of the unit root tests have estimation and testing problems. Hence the test of hypothesis becomes invalid and gives serious
misleading conclusions about the statistical significance of the estimated coefficients as well as the model. There are different
types of unit root tests for testing unit root. But these exiting tests have a number of drawbacks. Our main objective is to develop a
distance-based one-sided testing approach for testing unit root in different unit root tests. The usual method of testing unit root is
DF test considering three models. As the coefficient of is lies between 2 to 0, we found that our proposed DF test for testing
unit root gives better result than the usual DF test.
References
1. Dickey D.A., and Fuller W.A., Distribution of the Estimators for Autoregressive Time Series with a Unit Root, Journal of the
American Statistical Association,74, 427-431(1976)
2. Dickey D.A. and Fuller W.A., Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root, Econometrica, 49,
1057-1072 (1981)
3. Kwiatkowski D., Phillips P.C.B., Schmidt P. and Shin Y., Testing the Null Hypothesis of Stationary Against the Alternative of
a Unit Root, Journal of Econometrics, 54, 159-178 (1992)
4. Dickey D.A., Hasza D.P. and Fuller W.A., Testing for Unit Roots in Seasonal Time Series, Journal of the American Statistical
Association, 79, 355-367 (1984)
5. Dickey D.A. and Gonzalez-Farias G., A New Maximum Likelihood Approach to Testing for Unit Roots, unpublished paper
presented at the 1992 Joint Statistical meetings, August 9-13, Boston (1992)
6. Majumder A.K., one-Sided and Partially One-Sided Hypothesis Testing in Economics, unpublished PhD Thesis, Monash
University, Australia (1999)
7. Enders W., Applied econometric time series, John Wiley and Sons, New York (1995)
8. Dickey D.A., Estimation and Hypothesis Testing foe Nonstationary Time Series, Ph.D. Thesis, lowa Stat University, Ames,
lowa (1976)
9. DeJong D.N., Nankervis J.C., Savin N.E. and Whiteman C.H., The Power Problems of Unit Root Tests in Time Series with
Autoregressive Errors, Journal of Econometrics, 53, 323-343 (1992)
10. DeJong D.N. and Whiteman C.H., Reconsidering Trends and Random Walks in Macroeconomic Time Series, Journal of
Monetary Economics, 28 (1991)
11. Fuller W.A., Introduction to Statistical Time Series, New York John Wiley (1976)
12. Evans G. and Savin N.E., Testing for Unit Roots: I, Econometrica, 49, 753-779 (1981)
13. Evans G. and Savin N.E., Testing for Unit Roots: II, Econometrica, 52, 1241-1269 (1984)
14. Schwert G.W., Tests for Unit Roots: A Monte Carlo Investigation, Journal of Business and Economic Statistics, 7, 147-159
(1989)
15. Greene W.H., Econometric Analysis, 3rd ed., Prentice Hall, Upper Saddle River, New Jersey (1997)
16. Gujarati D.N., Basic Econometrics, 4th ed., McGraw Hill, New York (2003)
17. Hamilton J.D., Time Series Analysis, Princeton N.J,: Princeton University Press (1994)
18. Judge G.G., Griffiths W.E., Hill R.C., Lutkephol H. and Lee T.C., The Theory and Practice of Econometrics, 2nd ed., John
Willey and Sons, New York (1985)
19. Dickey D.A., Bell W.R. and Miller R.B., Unit roots series models: Tests and implications, Am, Statistical 40, 12-26 (1986)
20. Maddala G.S., Introduction to Econometrics, 2nd ed., Maxwell Macmillan, New York (1997)